# aijelen.hu > Track how AI models cite brands. Audit hallucinations, measure visibility, and fix reconstruction errors in ChatGPT, Perplexity, and Google AI Overviews. This site measures how AI models describe and cite companies. It treats citability as an infrastructure problem rather than a marketing challenge. ## Author - Zoltán Varga — AI visibility researcher and AI ecosystem architect, Budapest: https://vargazoltan.ai/ - English profile: https://zoltanvarga.ai/en/about/ - ORCID: https://orcid.org/0009-0003-1020-834X - arXiv author: https://arxiv.org/a/varga_z_1.html - Zenodo: https://zenodo.org/search?q=owners%3A1653886 ## Publications - Per-Entity Bias Mapping for AI Visibility: Why Brand Mentions Require Entity-Specific Calibration (arXiv:2606.21595): https://arxiv.org/abs/2606.21595 - AI Visibility in the Generative Era: A Manifesto for Entity-Level Measurement (DOI 10.5281/zenodo.20285424): https://doi.org/10.5281/zenodo.20285424 - Quality Challenges Do Not Escalate Source Fabrication in Large Language Models (RICE-P1) (DOI 10.5281/zenodo.20474050): https://doi.org/10.5281/zenodo.20474050 - Rejection-Induced Confabulation Escalation — Pilot Study 2 (RICE-P2) (DOI 10.5281/zenodo.20509755): https://doi.org/10.5281/zenodo.20509755 ## Key topics - AI visibility - brand hallucination - AI audit - GEO optimization - AEO strategy - LLMO - AI citations - synthetic media - brand presence AI - ChatGPT audit - Perplexity accuracy - Google AI Overviews - AI visibility index - AI láthatóság - márkahallucináció - AI auditálás - AI hivatkozás - generatív AI ## Questions and answers ### How do you measure AI visibility? We run repeated queries across multiple models to map output distributions. The audit analyzes raw generation, app processing, and user display layers to calculate fabrication rates and citation accuracy. ### What is the PEBM study? PEBM stands for Per-Entity Bias Map and analyzed 1,400 runs on 100 Hungarian B2B companies. It found that 45.9% of cited sources were fabricated overall, with higher rates for well-known firms. ### Do AI models rank experts like Google? No, AI systems reconstruct experts based on available data rather than ranking them. Citability depends on whether a model can point to a machine-readable, canonically anchored entity. ### Why do famous brands get wrong citations? The brand hallucination paradox suggests familiarity gives models more material for plausible but unsupported completions. Well-known brands may receive confident but false attributions due to this effect. ### Is SEO enough for AI visibility? SEO targets deterministic search algorithms while AI visibility addresses stochastic synthesis. Optimizing for AI requires making content easy to quote and structurally sound for model ingestion. ### What costs are involved in an audit? Costs depend on the number of queries and models tested, such as the 650-answer crop-protection study. Specific pricing varies based on the depth of the 25-dimension audit required. ## Related - AI visibility expert page: https://vargazoltan.ai/ai-lathatosagi-szakerto/ - AI visibility audit: https://vargazoltan.ai/audit/ - Research: https://vargazoltan.ai/kutatas/ ## Articles - [AI Visibility Is a Distribution: How Many Runs You Need](https://aijelen.hu/articles/visibility-is-a-distribution/): To measure AI visibility reliably, you must ask the same question at least 30 times to capture the underlying probability distribution rather than relying on a single anecdotal result. This sample size allows you to calculate a meaningful mention rate and identify structural variability caused by server routing, system prompts, or active testing layers within the AI platform. - [Why AI Keeps Repeating Outdated Company Facts](https://aijelen.hu/articles/parametric-retrieval-lag/): AI assistants repeat outdated company facts because they rely on two conflicting knowledge layers: retrieval systems that update in days and parametric memory that updates only every twelve to twenty-four months. This lag asymmetry allows old and new information to coexist in answers, creating a persistent gap between your current reality and the model's internal state. - [How Model Updates Break Your Visibility Trend Lines](https://aijelen.hu/articles/model-updates-break-trend-lines/): AI model updates introduce structural breaks in visibility data, rendering single-point measurements obsolete and invalidating direct comparisons with past baselines. To track brand presence accurately, you must treat every major model version as a distinct dataset, using distribution-based sampling rather than point estimates to establish comparable trend lines. - [Did Visibility Change Because of You or the Model?](https://aijelen.hu/articles/attributing-visibility-changes/): To determine if an AI visibility change stems from your actions or the model, you must log every intervention, run parallel control queries against a rival entity, and apply difference-in-differences analysis over a 90-day period. This method separates your specific impact from background noise like model updates or platform-wide traffic shifts. Without this instrumentation, you cannot distinguish between genuine optimization results and random stochastic variation. - [Which AI Visibility Gains Last and Which Fade?](https://aijelen.hu/articles/which-visibility-gains-last/): Improvements in AI visibility often fade when platforms adjust their safety filters or update training data, whereas gains built on third-party authority and knowledge graphs tend to persist. Short-term optimization tactics like keyword stuffing or schema markup rarely survive model retraining cycles without underlying brand credibility. Sustainable visibility requires continuous tracking of citation patterns over quarters rather than relying on one-off snapshots. - [Checklist: What to Log for Every Tracked AI Query](https://aijelen.hu/articles/ai-query-tracking-log-checklist/): To track a brand's presence in AI answers reliably, you must record the exact prompt, platform name, model version, timestamp, and full raw output including any source panels. You also need to log the specific query class, the extracted claim units, and the human verification label for each claim to enable statistical analysis and change detection over time. - [Raw Mention vs Verified Mention: A Deep Dive into AI Visibility](https://aijelen.hu/articles/raw-versus-verified-mention/): A raw mention occurs whenever an AI system displays an entity's name, regardless of accuracy or context. A verified mention requires that the entity appears alongside fact-checked, source-supported claims that align with reality. Counting raw mentions overstates presence, while verified mentions reveal the true trend of brand representation. - [The Four Visibility States a Brand Can Be In](https://aijelen.hu/articles/four-visibility-states/): A brand exists in one of four visibility states within AI answers: robust, fragile, systematic absence, or random presence. These states are defined by the average appearance rate and the variance across different query contexts and platforms. Tracking movement between these states over time provides actionable intelligence that a single snapshot metric cannot offer. - [What a Recurring AI Presence Report Should Contain](https://aijelen.hu/articles/ai-presence-report-structure/): A recurring AI presence report must include a stable query panel tested against a fixed baseline, a verified-fact rate measuring accuracy versus hallucinations, and a delta analysis showing change over time. It also requires a lag profile comparing parametric memory against retrieval updates and a list of open risks requiring immediate remediation. - [AI Presence Tracking FAQ: Frequency, Sampling, and Change Detection](https://aijelen.hu/articles/ai-presence-tracking-faq/): Track brand presence in AI answers by running stratified sampling across diverse query types and platforms, repeating measurements to capture statistical variance rather than relying on single snapshots. Meaningful change is identified only when shifts exceed the baseline standard deviation across a golden dataset of reference questions.